Pick a budget and a health-effects anchor. See what the current evidence supports.

The analyses answer two different questions over two different horizons. The first-year screen asks whether one year of monetized health benefit covers the upfront program cost. The 30-year decision analysis asks which portfolio has the best discounted value and remains preferable under uncertainty. Portfolio D is unfavorable in the first-year screen but becomes the regret-minimizing choice over the 30-year decision horizon. Those are different questions, not competing versions of the same number. A later six-axis joint stress test is not shown here: its upstream provenance became stale, so that result has been withdrawn pending a clean cascade rerun. Switch to the Krewski 2009 or Di 2017 anchors and some cells gray out — we don’t disaggregate per-portfolio net benefit at those anchors. What’s available is shown; nothing is fabricated.

Showing the multi-pollutant posterior from Investigation 6-3 (HR 1.28, β=0.02439). Krewski 2009 and Di 2017 per-portfolio NB is not disaggregated in the cascade — select them to see what data exists.

First-year screen Immediate affordability Mean NB and P(NB>0), sourced from Investigations 15/23.

30-year decision Discounted portfolio value P(optimal) from Investigation 44. The later joint stress test is withheld.

Loading portfolio data…

First-year columns are a one-year benefit-versus-upfront-cost screen. The 30-year columns apply discounted recurring benefits. Use the horizon that matches the decision; do not compare their probability values as though they came from one model run.

What the numbers do and do not show

The runner-up (Portfolio Q_nsga_2, the NSGA-II Pareto dominator) wins five of five robustness lenses, but its real-world deaths-avoided verdict is AMBIGUOUS. Under the Di health-effects anchor it delivers 267 deaths/yr (below our confidence threshold); under Krewski it delivers 895 (passes). Indoor air quality dose-response uncertainty spans 3.4×, so the call depends on which anchor you trust. Until that’s resolved, the conservative fallback is the zero-cost Portfolio A (the free-lunch baseline).
Portfolio D (the $4 B all-in option) is the regret-minimizing choice under our multi-pollutant health-effects estimate. P(optimal) = 1.000 across 8,192 quasi-Monte Carlo draws. The later six-axis confirmation is withdrawn because its upstream provenance is stale. Portfolio D wasn’t evaluated in the same robustness ladder as the runner-up — that ladder was built on a different portfolio set.
The health-effects anchor matters enormously. Under our multi-pollutant posterior (HR 1.28), every lens points to either Portfolio D or the runner-up Q_nsga_2. Under the Krewski 2009 (HR 1.06) or Di 2017 (HR 1.07) anchors, we don’t disaggregate per-portfolio net benefit at the same resolution — those cells show “—”, not invented numbers.
At the smaller $300 M budget the winner changes. At that scale the regret-minimizing choice is Q3-B (cooking-stove retrofits with the free-lunch baseline), not Portfolio D. Broader portfolios win at larger budgets because more sectors are affordable; concentrated, high-marginal-value interventions win at small budgets.
The joint stress test is withheld. Its historical page remains available as a clearly marked audit snapshot, but none of its probabilities or driver rankings are current evidence. A clean rerun through the dependency and freshness gates is required before they return here.
Numbers come from the 54 current investigations. Three withdrawn records remain available as audit history but do not support this dashboard.